amd/RyzenAI-SW

Ryzen AI Software is a toolkit for optimizing and deploying AI inference on AMD Ryzen AI powered PCs.

The software addresses the challenge of efficiently running AI models on consumer-grade Ryzen AI processors by providing tools and runtime libraries that handle model optimization and deployment. It works by integrating with ONNX Runtime through the Vitis-AI execution provider, allowing developers to take trained models and adapt them for the NPU and GPU acceleration available on Ryzen AI systems. The approach emphasizes ease of use alongside performance, enabling developers to leverage hardware acceleration without deep expertise in low-level optimization.

Developers building AI applications for Ryzen AI PCs should consider this toolkit as their primary path to hardware acceleration. It suits projects ranging from computer vision tasks like image classification and object detection to large language model inference and vision-language models. The repository provides getting started tutorials with ResNet models, examples covering various architectures including DistilBERT and YOLOv8, and demonstrations of advanced patterns like NPU-GPU pipelines. Windows ML integration is also available for developers preferring that framework. The toolkit includes quantization support through AMD Quark and handles both standard precision and BF16 model formats.

The project maintains a structured collection of working examples across multiple AI domains, with each example including its own documentation and setup instructions. The repository uses Git Large File Storage to manage model weights and demonstration assets, requiring developers to configure LFS before cloning. The codebase spans CNN examples, transformer-based models, large language model inference patterns, and specialized applications like automatic speech recognition and optical character recognition, indicating active development across diverse inference scenarios.